Learning to See in the Dark

Slides:



Advertisements
Similar presentations
Classification spotlights
Advertisements

ISO, Aperture and Shutter Speed For Beginners. The photographer can control how much natural light reaches the sensor by adjusting the camera's ISO shutter.
Digital Photography with Flash and No-Flash Image Pairs By: Georg PetschniggManeesh Agrawala Hugues HoppeRichard Szeliski Michael CohenKentaro Toyama,
Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節.
1 Invariant Local Feature for Object Recognition Presented by Wyman 2/05/2006.
Dynamic Range Compression & Color Constancy Democritus University of Thrace
ECE 472/572 – Digital Image Processing Lecture 2 – Elements of Visual Perception and Image Formation 08/25/11.
Detecting Electrons: CCD vs Film Practical CryoEM Course July 26, 2005 Christopher Booth.
Automatic Compensation for Camera Settings for Images Taken under Different Illuminants School of Electrical Engineering and Computer Science Kyungpook.
Mutual Information-based Stereo Matching Combined with SIFT Descriptor in Log-chromaticity Color Space Yong Seok Heo, Kyoung Mu Lee, and Sang Uk Lee.
KinectFusion : Real-Time Dense Surface Mapping and Tracking IEEE International Symposium on Mixed and Augmented Reality 2011 Science and Technology Proceedings.
Accidental pinhole and pinspeck cameras: revealing the scene outside the picture A. Torralba and W. T. Freeman Proceedings of 25 IEEE Conference on Computer.
Image Input Image acquisition or digitization. Sources include- Cameras Video camera w/ framegrabber Digital CCD camera w/ interface card Color or Monochrome.
Presented by Daniel Khashabi Joint work with Sebastian Nowozin, Jeremy Jancsary, Andrew W. Fitzgibbon and Bruce Lindbloom.
The Reason Tone Curves Are The Way They Are. Tone Curves in a common imaging chain.
Machine Vision Introduction to Using Cognex DVT Intellect.
A NOVEL METHOD FOR COLOR FACE RECOGNITION USING KNN CLASSIFIER
EE 638: Principles of Digital Color Imaging Systems Lecture 17: Digital Camera Characterization and Calibration.
Chapter 4 Exposure The Exposure Triangle and ISO.
Learning Photographic Global Tonal Adjustment with a Database of Input / Output Image Pairs.
U Fast Shutter Speed = Stops the Action u Slow Shutter Speed = Blurs the Action (Dragging the shutter) 1/6th 1/500th Photography Basics u Aperture and.
Camera Operations Pentax K1000. #2 & 3: Shutter Speed Dial/Index controls exposure time Slow shutter speed = longer exposure time = more blur and.
Philipp Gysel ECE Department University of California, Davis
Learning to Compare Image Patches via Convolutional Neural Networks SERGEY ZAGORUYKO & NIKOS KOMODAKIS.
When deep learning meets object detection: Introduction to two technologies: SSD and YOLO Wenchi Ma.
Deeply-Recursive Convolutional Network for Image Super-Resolution
Heechul Han and Kwanghoon Sohn
O BOSCH owns the night Our focus is to make darkness totally visible
The Relationship between Deep Learning and Brain Function
Object Detection based on Segment Masks
ISBI Camelyon16 Challenge Prague, April 13, 2016
Cascade of multi-scale convolutional neural networks for bone suppression of chest radiographs in gradient domain  Yang Wei , Chen Yingyin , Liu Yunbi.
Convolutional Neural Fabrics by Shreyas Saxena, Jakob Verbeek
Presenter: Ibrahim A. Zedan
Jure Zbontar, Yann LeCun
A Neural Approach to Blind Motion Deblurring
Multiplexed Illumination
Aperture.
Ajita Rattani and Reza Derakhshani,
Synthesis of X-ray Projections via Deep Learning
Summary Presentation.
CS6890 Deep Learning Weizhen Cai
R-CNN region By Ilia Iofedov 11/11/2018 BGU, DNN course 2016.
A Convolutional Neural Network Cascade For Face Detection
Bird-species Recognition Using Convolutional Neural Network
Introduction to Neural Networks
DEPTH RANGE ACCURACY FOR PLENOPTIC CAMERAS
Rob Fergus Computer Vision
إستراتيجيات ونماذج التقويم
RGB-D Image for Scene Recognition by Jiaqi Guo
(super) quick DSLR Camera guide
Single Image Rolling Shutter Distortion Correction
Creating Data Representations
KFC: Keypoints, Features and Correspondences
Image to Image Translation using GANs
Camera Basics Digital Photography.
Deep Neural Networks for Onboard Intelligence
Convolutional Network by GoogLeNet
Progress Report 2019/5/5 PHHung.
Image Processing and Multi-domain Translation
VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
Unrolling the shutter: CNN to correct motion distortions
Multi-UAV to UAV Tracking
Week 3 Volodymyr Bobyr.
Single Parameter Tuning
Introduction Face detection and alignment are essential to many applications such as face recognition, facial expression recognition, age identification,
Computing the Stereo Matching Cost with a Convolutional Neural Network
Nguyen Ngoc Hoang, Guee-Sang Lee, Soo-Hyung Kim, Hyung-Jeong Yang
Directional Occlusion with Neural Network
Presentation transcript:

Learning to See in the Dark Published at the Conference on Computer Vision and Pattern Recognition (CVPR 2018) Presented by: Nafiseh Vahabi

Challenges of fast imaging in low light Low photon count Low SNR Short-exposure images suffer from noise Long exposure can induce blur and is often impractical. This challenge is addressed by: Introduce a dataset of raw short-exposure low-light images, with corresponding long-exposure reference images Develop a pipeline for processing low-light images, based on end-to-end training of a fully- convolutional network.

Camera Setting The network operates directly on raw sensor data Camera setting : (Two cameras) Illumination < 0.1 lux at the camera Exposure time : 1/30 Aperture : f/5.6

SID (See in the Dark) Dataset Dataset contains 5094 raw short-exposure images, each with the reference long-exposure image.

Example of Dataset 0.2 < outdoor illumination < 5

Existing Methods

ConvNet Structure Two structure: Multi-scale context aggregation network U-net (default ) Memory consumption (full resolution image in GPU) The amplification ratio determined the brightness of output (set externally)

Result of Two Structures

Training Training from scratch using L1 and Adam optimizer Input: raw data, short exposure image Ground Truth: Corresponding long exposure image in RGB Network trained for each camera Amplification ratio is set to be a difference between the input and reference images Learning rate: 0.0001 Training proceeds to 4000 epochs

Evaluate the Proposed Methods

Evaluate the Proposed Methods

Conclusion Successful noise suppression Correct color transformation Limitation No humans or dynamic objects The result is imperfect and need to improve Application ratio set externally Improve runtime optimisation